Active Magnetic Bearing Tuning Ability in Reasoning Large Language Models

Active Magnetic bearing (AMB) is widely applied to high-speed rotation machines because of its advantages of no contact and no friction. However, the AMB is unstable without a tuned displacements controller and the tuning process greatly depends on the engineering experience or the accurate models, causing the high use-cost. The large language models (LLMs) develop recently and can help solve some practical problems. The applications of LLMs in industry fields are also more common. In this article, the reasoning LLMs including the Deepseek-r1 and the ChatGPTo3-mini are used to help tune the PID controller parameters of a single axis AMB system. The first test is telling the PD parameters' size relation compared to the suitable PD parameters when the AMB is not stable, which relies on the LLMs' knowledge. The second test is the continuous PD parameters tuning to get a better performance, which relies on the LLMs' reasoning ability. According to the test results, the tested LLMs have little prior knowledge about the AMBs, making them difficult to give the correct tuning advice directly. Nevertheless, in the continuous tuning, the reasoning LLMs can study from the changes in the waveforms of different PD parameters and provide true tuning direction gradually, like human beings. If the additional AMB knowledge is supplied initially, the tested LLMs performance better. The test shows the possibility of the preliminary application in the AMB system and the fine-tuning is necessary for better performance.